An Abstract Bohm An Abstract Bohm NormalizationNormalization

John Glauert & zurab Khasidashvili

(1) UEA, UK; (2) BIU&Intel, Israel

OverviewOverview

Normalization by needednessRelative normalizationAbstract relative normalizationInfinite results of finite termsStability, regularity and superstabilityBohm normalization and minimality resultsConclusions

Normalization by needednessNormalization by neededness

Developed by Huet & Levy, 1979.In an orthogonal TRS, a redex in term t is

needed if its residual is constructed in every normalizing reduction of t.

If t has a normal form, it can be found by repeatedly contracting needed redexes

Normalization wrt other sets of Normalization wrt other sets of normal formsnormal forms

Barendregt et al. studied normalization wrt head-normal forms, in the calculus

Maranget - weak-head-normal formsNocker – constructor head-normal formsMiddeldorp – root-stable formsGlauert & Khasidashvili formalized a well-

behaved concept of `partial results’

Relative neededness &Relative neededness &Stable sets of resultsStable sets of results

Let S be a set of terms. A redex u in term t is S-needed if any S-normalizing reduction starting from t contracts a residual of u.

A set S of terms is stable if:– It is closed under reduction– Every step entering S is S-needed

A non-stable setA non-stable set

A non-stable set: Set S={I(x),x}Reduction relation: R ={I(x) x}Term I(I(x)) has no S-needed redex; it has

an S-normal form (actually, it has two).

I(x)

I(I(x))

x

S

Normalization wrt a reductionNormalization wrt a reduction

Let P: t…s be a reduction A redex u in t is P-needed if it is contracted in any

reduction Levy-equivalent to P. P is self-needed or standard if it only contracts P-

needed redexes. For `regular’ reductions P, we have shown that by

contracting P-needed P-erased redexes we can built standard reductions Levy-equivalent to P.

Levy-equivalenceLevy-equivalence

Levy-equivalence on finite co-initial reductions is generated by the axioms:– U+V/U=LV+V/U

– P=P’ N+P+Q =L N+P’+Q

where U (resp. V) denotes complete development of redex set U, and U/V denotes the set of residuals of redexes in U after performing V, as well as the corresponding complete development.

Levy-equivalence (cont.)Levy-equivalence (cont.)

Klop diagram: P =L Q P/Q=Q/P=0

. . .. .

. . .. .

. . .. .

. . .. . ** ****

** **

*

***

* * * * *

**

****

P

Q/P

P/Q

Q

Deterministic Residual Deterministic Residual Structures (DRS)Structures (DRS)

A DRS consists of– An Abstract Reduction System A=(Ter,Red,)

Ter is set of objects, called termsRed is set of redexes (or redex occurrences) associates to every redex its source and target

terms; redexes are written u : t s A term may have a finite number of redexes

– A residual relation, denoted /, between redexes in the source and target terms of .

Deterministic Residual Deterministic Residual Structures (cont.)Structures (cont.)

/ satisfies three `permutation’ axioms:– If t s, then a redex u in s may be a residual of

at most one redex v in t; otherwise u is created;– u/u = 0 (empty reduction);– All developments terminate; all complete

developments of a set U of redexes in t end at the same term; and residuals of a redex v in t under all complete developments of U are the same.

Deterministic Residual Deterministic Residual Structures (cont.)Structures (cont.)

The residual relation extends to finite reductions by transitivity, and Levy equivalence can be defined.

/ satisfies an `advanced’ axiom:– [weak acyclicity] (E.Stark [Sta89])

u/v=0 & !u=v !v/u=0

Stable DRSsStable DRSs

A DRS is stable if in addition the following axiom is satisfied:– [stability] (modification of [GLM’92])

!u=v & u creates w v/u creates w/(v/u)

.

. .

.u

u/v

v *

w

w/(v/u)

v/u**

-embedding-embedding

Define P Q iff P/Q=0.Theorem: Let be a set of reductions

starting from t. Then -meet of reductions in can be computed as follows: – U + (/U)

where U is the set of all redexes t such that: U

L

L

L

L

L L

L

Relative needednessRelative neededness

Let S be a set of reductions in a DRSWe call a redex u in t S-unneeded if there is

a Q in S that is external to u (i.e., does not contract residuals of u), and is S-needed otherwise.

A reduction with S-(un)needed steps is S-(un)needed

Stable ordering: Stable ordering:

Let S be a set of reductions in a DRSIf P and Q are finite, define P Q iff P/Q

is S-unneeded.If P and/or Q are infinite, define P Q iff

for any initial part P’ of P there is an initial part Q’ of Q such that P’ Q’.

S-equivalence: P =S Q iff P Q & Q P.S

S

S

S

S S

Stable sets of reductionsStable sets of reductions

A set S of reductions is called stable iff:– !P’ in S & P’+P” in S P” in S– P is S & P Q Q in S– Every non-empty P in S contracts at least one

S-needed redex

S

Regular and superstable setsRegular and superstable sets

A set S of reductions is called regular iff:– In no term can an S-unneeded redex duplicate

and S-needed redex

S is called superstable iff:– For any S-normalizable term t, S contains a

unique, up to =L, -minimal reduction starting from t. Such reductions are called S-minimal.

L

Bohm NormalizationBohm Normalization

P : t0 t1 … is S-needed fair if for any S-needed redex vi in ti, vi Pi, where Pi is the suffix of P starting from ti.

A redex u in t is S-erased if it does not have a residual under any S-normalizing reduction starting from t.

S

Bohm Normalization (cont.)Bohm Normalization (cont.)

Theorem: Let S be a regular set of reductions, and let t be S-normalizable.– Any S-needed fair reduction starting from t is

S-normalizing.– If S contains a finite reduction starting from t,

then t does not have a reduction in which infinitely many times S-needed redexes are contracted.

S

A characterization of A characterization of superstabilitysuperstability

Theorem: Let S be a regular stable set of terms, in an SDRS. Then S is superstable iff any S-normalizable term t not in S contains an S-erased S-needed redex.

Minimal relative normalizationMinimal relative normalization

Theorem: Let S be a superstable set of terms in an SDRS, and let T be S-normalizable term not in S. S-minimal S-normalizing reductions arise from repeatedly contracting S-needed S-erased redexes. A finite number of S-unneeded but S-erased redexes may also be contracted without loosing S-minimality.

Related workRelated work

On Abstract reductions with residuals:– Started by Stark, and by Gonthier, Levy and

Mellies; further developed by Mellies.

On minimal reductions:– Our early minimality results (for orthogonal

ERSs) were inspired by Maranget.– Mellies later proved a minimality result by

using a nesting axioms.

ConclusionsConclusions

We have unified our earlier relative and discrete normalization results.

We use this abstract framework in a number of articles to study operational, denotational as well as event based semantics of orthogonal rewrite systems.

Generalizing these results to infinite terms (with infinitely many redexes) requires enreaching SDRSs with axioms studied by Kennaway (CWI report).